随着道路上车辆的增多与人们对高质量生活的需求，协助交警对危险驾驶行为进行实时检测至关重要。目前主要利用卷积神经网络进行驾驶员接打手持电话、发短信等危险驾驶行为检测，解决了传统方法的局限性以及准确度较低的问题，但仍存在网络参数多、数据集背景噪声大、对手机目标检测精确度低且定位不准确等问题。论文基于卷积神经网络方法对危险驾驶行为检测问题进行研究，主要研究工作如下：

（1）针对目前危险驾驶行为检测方法存在精确度不高且实时性差的问题，将YOLOv5应用于危险驾驶行为检测这个复杂环境中。另外针对YOLOv5训练过程中背景噪声大导致目标特征不显著造成边界框回归损失较大，以及模型参数较多这两个问题，提出一种基于注意力机制的轻量化危险驾驶行为检测方法。首先将CBAM引入到特征提取网络，其次为了降低网络参数量同时避免引入模块影响检测速度，用Ghost卷积代替普通卷积。最后在公开数据集上进行实验，实验结果表明，该算法不仅降低了边界框回归损失，还加快了推理速度，降低了网络参数量，使模型更加轻量化。

（2）针对上述危险驾驶行为检测方法检测驾驶员接打手持电话时会存在误判问题，先提出两阶段判断流程来识别该行为。然后针对YOLOv5对手机目标尤其是当驾驶员左手手持手机检测时存在精确度低、定位不准确的问题，通过在特征融合阶段采用BiFPN，并融合多层CA模块，提出一种基于特征融合的驾驶员接打手持电话行为检测方法。实验结果证明，该算法的准确度达到99.8%，提高了驾驶员左手手持手机检测的精确度，能更加准确定位到左手手持手机特征，还缩短了推理时间。

（3）结合论文提出的危险驾驶行为检测算法，设计与实现了C/S架构的All平安检测系统。该系统包括输入与模型选择、检测驾驶员接打电话、向后够东西、操作收音机等危险驾驶行为、结果统计与保存等功能。最后对系统进行功能与非功能测试，测试结果表明，该系统基本符合预期，对协助交警检测危险驾驶行为具备一定的实用性。

With the increase of vehicles on the road and people’s demand for a high quality of life, it is crucial to assist traffic police in real-time detection of dangerous driving behaviors. At present, the convolutional neural network is mainly used to detect dangerous driving behaviors, such as answering a hand-held phone and sending text messages, which resolves the limitations and low accuracy of traditional methods. However, there are still some problems, such as multiple network parameters, large background noise of data set, low accuracy and inaccurate location of mobile phone target detection. Based on convolutional neural network method, this paper studies the detection of dangerous driving behavior. The main research work is as follows:

(1) Aiming at the problems of low accuracy and poor real-time performance of current dangerous driving behavior detection methods, YOLOv5 is applied to the complex environment of dangerous driving behavior detection. In addition, in order to solve the problems of large background noise in the training process of YOLOv5, which leads to inapparent target features and large bounding box regression loss, and the large number of model parameters, a lightweight dangerous driving behavior detection method based on attention mechanism is proposed. Firstly, CBAM is introduced into the feature extraction network. Secondly, in order to reduce the number of network parameters and avoid the introduction of modules affecting the detection speed, Ghost convolution is used to replace ordinary convolution. Finally, the experimental results on the public data set show that the proposed algorithm not only reduces the bounding box regression loss, but also accelerates the inference speed, reduces the number of network parameters, and makes the model more lightweight.

(2) Aiming at the problem of misjudgment when the above dangerous driving behavior detection method detects the driver’s behavior of answering a handheld phone, a two-stage judgment process is proposed to identify the behavior. Then, aiming at the problems of low accuracy and inaccurate location of YOLOv5 when detecting the mobile phone target, especially when the driver holds the mobile phone in his left hand. By using BiFPN in the feature fusion stage and integrating the multi-layer CA module, a driver’s behavior detection method based on feature fusion for answering a handheld phone is proposed. Experimental results show that the accuracy of the algorithm reaches 99.8%, which improves the accuracy of the driver’s left hand mobile phone behavior detection, can more accurately locate the mobile phone’s features in the left hand, and also shorten the reasoning time.

(3) Combined with the dangerous driving behavior detection algorithm proposed in this paper, the All safety detection system based on C/S architecture is designed and implemented. The system includes the functions of input and model selection, detecting the driver’s dangerous driving behaviors such as making a phone call, reaching back, operating the radio and so on, and counting and saving the results. Finally, the functional and non-functional tests of the system are carried out. The results show that the system basically meets the expectations, and it has certain practicability to assist the traffic police to detect dangerous driving behavior.

[1]陈力维, 高润泽. 我国新能源汽车技术发展现状分析[J]. 交通节能与环保, 2021, 17(06): 14-19.

[2]Liu Z, Hao H, Cheng X, et al. Critical issues of energy efficient and new energy vehicles development in China[J]. Energy Policy, 2018, 115(01): 92-97.

[3]王克辉. 全国机动车保有量突破4亿辆[N]. 人民公安报, 2022-07-07(002).

[4]公安部办公厅统计处. 2021年全国及各省（区市）道路交通事故情况[J]. 公安研究, 2022, 333(07): 95-96.

[5]Zhao Z, Xia S, Xu X, et al. Driver Distraction Detection Method Based on Continuous Head Pose Estimation[J]. Computational Intelligence and Neuroscience, 2020, 2020(04): 1-10.

[6]Zhang L, Cui B, Yang M, et al. Effect of Using Mobile Phones on Driver’s Control Behavior Based on Naturalistic Driving Data[J]. International Journal of Environmental Research and Public Health, 2019, 16(08): 1464.1-1464.13.

[7]Hayley A C, Shiferaw B, Aitken B, et al. Driver monitoring systems (DMS): The future of impaired driving management?[J]. Traffic injury prevention, 2021, 22(04): 313-317.

[8]王丹. 基于机器视觉的驾驶员打电话行为检测[D]. 北京：北京理工大学, 2015.

[9]潘超鹏. 基于计算机视觉的驾驶员驾驶行为识别[D]. 湖南：湖南大学, 2021.

[10]Shahverdy M, Fathy M, Berangi R, et al. Driver behavior detection and classification using deep convolutional neural networks[J]. Expert Systems with Applications, 2020, 149(C): 113240.1-113240.12.

[11]李俊俊, 杨华民, 张澍裕, 等. 基于神经网络融合的司机违规行为识别[J]. 计算机应用与软件, 2018, 35(12): 222-227+319.

[12]卜庆志, 裘君, 胡超. 基于HOG特征提取与SVM驾驶员注意力分散行为检测方法研究[J]. 集成技术, 2019, 8(04): 69-75.

[13]Zhang L, Tan B, Liu T, et al. Research on Recognition of Dangerous Driving Behavior Based on Support Vector Machine[C]//2020 12th International Conference on Graphics and Image Processing (ICGIP). SPIE, 2021: 11720L.1-11720L.6.

[14]Tran D, Manh Do H, Sheng W, et al. Real‐time detection of distracted driving based on deep learning[J]. IET Intelligent Transport Systems, 2018, 12(10): 1210-1219.

[15]Al-Hussein W A, Por L Y, Kiah M L M, et al. Driver Behavior Profiling and Recognition Using Deep-Learning Methods: In Accordance with Traffic Regulations and Experts Guidelines[J]. International journal of environmental research and public health, 2022, 19(03): 1470.1-1470.23.

[16]Abosaq H A, Ramzan M, Althobiani F, et al. Unusual Driver Behavior Detection in Videos Using Deep Learning Models[J]. Sensors, 2022, 23(01): 311.1-311.20.

[17]代少升, 黄向康, 黄涛, 等. 一种基于深度学习的驾驶员打电话行为检测方法[J]. 电讯技术, 2021, 61(07): 785-792.

[18]Wu W S, Lu Z M. A Real-Time Cup-Detection Method Based on YOLOv3 for Inventory Management[J]. Sensors, 2022, 22(18): 6956.1-6956.17.

[19]熊群芳, 林军, 岳伟, 等. 基于深度学习的驾驶员打电话行为检测方法[J]. 控制与信息技术, 2019, 2019(06): 53-56+62.

[20]Chen Z, Guo H, Yang J, et al. Fast vehicle detection algorithm in traffic scene based on improved SSD[J]. Measurement, 2022, 201: 111655.1-111655.11.

[21]文国波, 袁泉, 郭海涛, 等. 基于深度学习的乘务员接打电话行为检测方法研究[J]. 工业控制计算机, 2021, 34(03): 24-27.

[22]许婷婷, 傅俊琼, 罗昆. 基于CNN和多尺度融合的驾驶员打电话行为检测[J]. 计算机技术与发展, 2022, 32(02): 88-93.

[23]Shafiq M, Gu Z. Deep Residual Learning for Image Recognition: A Survey[J]. Applied Sciences, 2022, 12(18): 8972.1-8972.43.

[24]柳长源, 虎浩媛, 毕晓君. 双线性融合网络的驾驶员分心行为识别[J]. 北京邮电大学学报, 2022, 45(02): 79-84.

[25]许腾, 唐贵进, 刘清萍, 等. 基于空洞卷积和Focal Loss的改进YOLOv3算法[J]. 南京邮电大学学报(自然科学版), 2020, 40(06): 100-108.

[26]杜虓龙, 余华平. 基于改进Mobile Net-SSD网络的驾驶员分心行为检测[J]. 公路交通科技, 2022, 39(03): 160-166.

[27]Sandler M, Howard A, Zhu M, et al. MobileNetV2: Inverted Residuals and Linear Bottlenecks[C]//2018 31st IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2018: 4510-4520.

[28]何丽雯, 张锐驰. 基于深度学习的驾驶员分心行为识别[J]. 计算机与现代化, 2022, 322(06): 67-74.

[29]Li Z, Liu F, Yang W, et al. A Survey of Convolutional Neural Networks: Analysis, Applications, and Prospects[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 33(12): 6999-7019.

[30]张焕, 张庆, 于纪言. 卷积神经网络中激活函数的性质分析与改进[J]. 计算机仿真, 2022, 39(04): 328-334.

[31]Li S, Zhang S, Xue J, et al. Lightweight target detection for the field flat jujube based on improved YOLOv5[J]. Computers and Electronics in Agriculture, 2022, 202: 107391.1-107391.13.

[32]Rani S, Singh B K, Koundal D, et al. Localization of stroke lesion in MRI images using object detection techniques: A comprehensive review[J]. Neuroscience Informatics, 2022, 2(03): 100070.1-100070.8.

[33]Guo Y, Zhang J, Su P, et al. The Study of Locating Diseased Leaves Based on RPN in Complex Environment[J]. Journal of Physics: Conference Series, 2020, 1651(01): 012089.1-012089.7.

[34]Niu C, Li K. Traffic Light Detection and Recognition Method Based on YOLOv5s and AlexNet[J]. Applied Sciences, 2022, 12(21): 10808.1-10808.18.

[35]Ren K, Chen Z, Gu G, et al. Research on infrared small target segmentation algorithm based on improved mask R-CNN[J]. Optik, 2023, 272: 170334.1-170334.11.

[36]Humayun M, Ashfaq F, Jhanjhi N Z, et al. Traffic Management: Multi-scale Vehicle Detection in Varying Weather Conditions Using YOLOv4 and Spatial Pyramid Pooling Network[J]. Electronics, 2022, 11(17): 2748.1-2748.29.

[37]刘宇宸, 石刚, 崔青, 等. 改进MobileNetv3-YOLOv3交通标志牌检测算法[J]. 东北师大学报(自然科学版), 2022, 54(02): 53-60.

[38]Ren S, He K, Girshick R, et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(06): 1137-1149.

[39]Zhang D, Hu J, Li F, et al. Small Object Detection via Precise Region-Based Fully Convolutional Networks[J]. Computers, Materials & Continua, 2021, 69(02): 1503-1517.

[40]Fu R, He J, Liu G, et al. Fast Seismic Landslide Detection Based on Improved Mask R-CNN[J]. Remote Sensing, 2022, 14(16): 3928.1-3928.19.

[41]Meimetis D, Daramouskas I, Perikos I, et al. Real-time multiple object tracking using deep learning methods[J]. Neural Computing and Applications, 2023, 35(01): 89-118.

[42]王楷元, 韩晓红. 基于 You Only Look Once v2 优化算法的车辆实时检测[J]. 济南大学学报(自然科学版), 2020, 34(05): 443-449.

[43]Wang C Y, Liao H Y M, Wu Y H, et al. CSPNet: A New Backbone that can Enhance Learning Capability of CNN[C]//2020 33rd IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). IEEE, 2020: 390-391.

[44]Zhu Q, Zheng H, Wang Y, et al. Study on the Evaluation Method of Sound Phase Cloud Maps Based on an Improved YOLOv4 Algorithm[J]. Sensors, 2020, 20(15): 4314.1-4314.18.

[45]许德刚, 王露, 李凡. 深度学习的典型目标检测算法研究综述[J]. 计算机工程与应用, 2021, 57(08): 10-25.

[46]Zhang Y, Zhou W, Wang Y, et al. A real-time recognition method of static gesture based on DSSD[J]. Multimedia Tools and Applications, 2020, 79(25): 17445-17461.

[47]Khan N, Singh A V, Agrawal R. Enhanced Deep Learning Hybrid Model of CNN Based on Spatial Transformer Network for Facial Expression Recognition[J]. International Journal of Pattern Recognition and Artificial Intelligence, 2022, 36(14): 2252028.1-2252028.28.

[48]Hu J, Shen L, Sun G. Squeeze-and-Excitation Networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 42(08): 2011-2023.

[49]Li X, Wang W, Hu X, et al. Selective Kernel Networks[C]//2019 32nd IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2019: 510-519.

[50]Li Y H, Aslam M S, Harfiya L N, et al. Conditional Wasserstein Generative Adversarial Networks for Rebalancing Iris Image Datasets[J]. IEICE Transactions on Information and Systems, 2021, 104(09): 1450-1458.

[51]Li Z, Xi T, Zhang G, et al. AutoDet: Pyramid Network Architecture Search for Object Detection[J]. International Journal of Computer Vision, 2021, 129(04): 1087-1105.

[52]Zhang X, Wang X. Motion deblurring method based on Improved DeblurGAN[J]. Academic Journal of Computing & Information Science, 2018, 3(04): 102-109.

[53]Wang C, Zhong C. Adaptive Feature Pyramid Networks for Object Detection[J]. IEEE Access, 2021, 9: 107024-107032.

[54]Woo S, Park J, Lee J Y, et al. CBAM: Convolutional Block Attention Module[C]//2018 15th European Conference on Computer Vision (ECCV), 2018: 3-19.

[55]Han K, Wang Y, Tian Q, et al. GhostNet: More Features from Cheap Operations[C]//2020 33rd IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2020: 1580-1589.

[56]Liu S, Qi L, Qin H, et al. Path Aggregation Network for Instance Segmentation[C]//2018 31st IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2018: 8759-8768.

[57]Zheng Z, Wang P, Liu W, et al. Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression[C]//2020 34th AAAI Conference on Artificial Intelligence (AAAI). Association for the Advancement of Artificial Intelligence, 2020: 12993-13000.

[58]Tan M, Pang R, Le Q V. Efficientdet: Scalable and Efficient Object Detection[C]//2020 33rd IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2020: 10781-10790.

[59]Hou Q, Zhou D, Feng J. Coordinate Attention for Efficient Mobile Network Design[C]// 2021 34th IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2021: 13713-13722.

[60]贾可心, 马正华, 朱蓉, 等. 注意力机制改进轻量SSD模型的海面小目标检测[J]. 中国图象图形学报, 2022, 27(04): 1161-1175.